- Title
- Image recognition for machine fault detection based on convolutional neural network
- Creator
- Khairul Amirin Emar Azami Faculty of Engineering - School of Mechanical Engineering
- Creator
- Muhammad Firdaus Isham, supervisor
- Creator
- Universiti Teknologi Malaysia
- Creator
- Faculty of Engineering - School of Mechanical Engineering
- Subject
- Mechanical engineering
- Date
- 2022 Pub. Date
- Type
- Thesis
- Relation
- Degree
- Description
- Bearings are an essential component of rotary machinery. They are also available in a vast array of sizes and dimensions and may be installed in a number of configurations to fulfil various requirements. Therefore, it is important to make sure that the bearing is at optimum condition and replaced immediately if found faulty. Bearing failure has detrimental effects on production schedules and operations. Consequently, accomplishing bearing issue detection and diagnosis in advance ensures the safety and reliability of rotating equipment systems. With the fourth industrial revolution that is associated with automation, factories and assembly lines now focuses on changing the norm by implementing artificial intelligence (AI) and fully automated machines to replace man power where an assembly line can operate with less human intervention and controlled by a few people only. Therefore, this paper proposes the utilization of machine fault detection using image recognition based on Convolutional Neural Network (CNN). CNN are known as one of machine learning models to get widespread attention due to their excellent performance in image processing applications. The bearing’s vibration signals are acquired from a test rig with four different conditions. The conditions are healthy bearings, inner race defects, outer race defects and ball bearing defects. With each conditions are recorded in the datasets, the data will be converted into spectrogram images before feeding it to the CNN model. The performance of the CNN model is based on the comparison of two different models which is Model A and Model B. Model B is developed based on the performance of Model A where the tunings of hyper parameters will be implemented to improve its performance and the model with better performance is chosen to proceed as the proposed model. The result shows that the proposed model able to detect and classify the bearing faults up to 99.9% accuracy.
- Description
- Project Paper (Bachelor of Engineering (Mechanical)) - Universiti Teknologi Malaysia, 2022
- Format
- Unpublished, 67 p., Completion
- Language
- ENG
- Rights
- Closed Access, UTM, Complete
- Identifier
- vital:153717, valet-20231009-072533
- Contributor
- fahmimoksen
- Site
- Restricted Repository (Click here to Restricted Repository. Then, sign-in to download material. Login required.)
- Hits: 2
- Visitors: 2
- Downloads: 1
| Thumbnail | File | Description | Size | Format | |||
|---|---|---|---|---|---|---|---|
| View Details | ATTACHMENT01 | KhairulAmirinEmarAzamiKPSKM2022.pdf | 12 MB | Adobe Acrobat PDF | View Details |